measuring the facebook advertising ecosystem€¦ · • 47.4% ages 22-30 vs 32.5% on facebook •...

Measuring the Facebook Advertising Ecosystem

Athanasios Andreou, Márcio Silva, Fabrício Benevenuto, Oana Goga, Patrick Loiseau, Alan Mislove

Advertising in Facebook

Facebook is the second biggest advertising platform:

More than the GDP of Iceland or Bahrain!

!2

Advertising in Facebook

Facebook is the second biggest advertising platform:

More than the GDP of Iceland or Bahrain!

!2

Millions advertisers and billions of users:

Facebook ad platform issues constantly in the spotlight

!3

This talk

Who are the advertisers in Facebook?

How are they targeting users?

!4

Methodology

!5

AdAnalyst — A tool to help you make sense of your ads

Methodology

!5


Chrome and Firefox extension

Collects Facebook ads and related data from real users

Methodology

!6


Increase transparency for users:

Our Dataset

DATA-Worldwide:

• Disseminated across friends, colleagues, conferences etc

• 114 users / 89K ads / 22K advertisers

• 50 users from France, 16 from Germany, 16 from US

DATA-Brazil:

• Disseminated as part of a project to monitor the 2018 Brazilian presidential elections

• 508 users / 146K ads / 28K advertisers

• 495 users from Brazil!7


Our users

Dissemination strategy introduces some biases

!8

Our users


Younger educated men:

• 47.4% ages 22-30 vs 32.5% on Facebook

• 71.1% with tertiary education vs 35.9%

• 68.4% men vs 57%

!8

Our users


Younger educated men:

• 47.4% ages 22-30 vs 32.5% on Facebook

• 71.1% with tertiary education vs 35.9%

• 68.4% men vs 57%

Our users are overall similar to the global Facebook population

• More likely to travel frequently and know people that are expats

!8

This talk



!9

Advertisers’ identity

!10

How popular are advertisers?


!10

Popular:

Ordinary:

Niche:Face

book

Lik

es



!10

Popular:

Ordinary:

Niche:Face

book

Lik

es

How trustworthy are advertisers?



!10

Popular:

Ordinary:

Niche:Face

book

Lik

es

How trustworthy are advertisers?• advertisers can verify their account

• phone number or official document

• government issued photo id



!10

Popular:

Ordinary:

Niche:Face

book

Lik

es

How trustworthy are advertisers?• advertisers can verify their account

• phone number or official document

• government issued photo id

• 64% of advertisers are not verified

Popular 66,9 % 6,1 %Ordinary 10,3 % 12,6 %Niche 0,2 % 6,4 %


Advertisers’ categories

!11

Advertisers can self-report their categories

943 categories:

• Nonprofit Organization, Shopping & Retail

• Evangelical Church, Aquarium, Opera

We group them to 35 broader categories (IAB)

Sensitive Categories

Advertisers’ categories

!11

Advertisers can self-report their categories

943 categories:

• Nonprofit Organization, Shopping & Retail

• Evangelical Church, Aquarium, Opera

We group them to 35 broader categories (IAB)

Sensitive Categories

1.Food and Drink: 9.3%2.Style & Fashion: 8.5%3.Technology and Computing: 8.4%4.Community Organization: 8.2%5.Shopping: 6.7%6.News and Politics: 5.5%7.Travel: 4.6%8.Education: 4.4%9.Healthy Living: 4.2%10.Home & Garden: 3.6%

• Business and finance: 2%• Medical Health: 1.2%• Legal: 0.2%• Religion and Spirituality: 0.1%

This talk



!12

Many strategies available for targeting

!13

Attribute-based


!13

Attribute-based

PII-based


!13

Attribute-based

PII-based

Lookalike audiences


!14


!14

Retargeting


!14

Location-based

Retargeting


!14

Location-based

Social Neighborhood

Retargeting

How do we know how advertisers target users?

!15

Limitations of Facebook Explanations

Facebook ad explanations reveal information about the targeting

!16

Limitations of Facebook Explanations

Facebook ad explanations reveal information about the targeting

They have limitations:

• They show one attribute

• Demographics > Interests > PII-based > Behaviors

• Show the most prevalent attribute

!16

What targeting strategies do advertisers use?

!17

Attribute-based: 47% of adsAge/Gender/Location: 23%Lookalike Audiences: 18% Retargeting: 8%Social Neighborhood: 3%PII-based: 2%Location-based: 1%


!17


74% of users PII-based, 92% Lookalike audiences


!17


74% of users PII-based, 92% Lookalike audiences

Differences between North America and Europe:

• PII-based 6% vs 2%

• Lookalike audiences 30% vs 17%

What attributes do advertisers use?

!18

2.5K attributes:


!18

• 38% of attributes appear only in one ad:

‣ Pokemon Yellow, Adobe After Effects, Modernism

2.5K attributes:


Attributes not always aligned with nature of advertiser:

• Google: Hacker News, Digital Media, Masters degree

• Also Google: Women’s rights, US politics (very liberal), Politics and social issues

!18



2.5K attributes:


Attributes not always aligned with nature of advertiser:

• Google: Hacker News, Digital Media, Masters degree

• Also Google: Women’s rights, US politics (very liberal), Politics and social issues

591 attributes used by News and politics advertisers:

• Anti-fascism, LGBT community

!18



2.5K attributes:

Do advertisers change ads over attributes?

2.5K advertisers that have used more than one attributes

65% of them never used the same text with two different attributes

VICE news:

• PC Magazine: “A self-driving, flying taxi could soon be a reality”

• Democratic Party: “Mr. Trump and Mr. Cohen have a lot of explaining to do”

• US Politics (Very Liberal): “One of the reasons it’s hard for Trump to navigate the guns issue after Parkland is that the gun rights community itself is still trying to figure out what change is acceptable”

•

!19

Popular and verified advertisers mixed with niche and unverified

Advertisers that can be sensitive

They use strategies that can be considered invasive or opaque

The use a plethora of attributes that are difficult to monitor

They change the content of the ads w.r.t. attributes they use

Conclusion

!20

Popular and verified advertisers mixed with niche and unverified

Advertisers that can be sensitive

They use strategies that can be considered invasive or opaque

The use a plethora of attributes that are difficult to monitor

They change the content of the ads w.r.t. attributes they use

Conclusion

!20

Need for better mechanisms to audit ads and advertisers

Need for global approach towards transparency

A tool to help you make sense of your Facebook ads

https://adanalyst.mpi-sws.org

Disable adblockers for Facebook and anti-tracking

Set your Facebook in English or French

AdAnalyst

!21

https://adanalyst.mpi-sws.org

QUESTIONS

!22

Capturing Facebook Ads

!23

measuring the facebook advertising ecosystem€¦ · • 47.4% ages 22-30 vs 32.5% on facebook •...

Documents